Modeling Intermolecular and Intramolecular Modes of Liquid Water Using Multiple Heat Baths: Machine Learning Approach.

The vibrational motion of molecules in dissipative environments, such as solvation and protein molecules, is composed of contributions from both intermolecular and intramolecular modes. The existence of these collective modes introduces difficulty into quantum simulations of chemical and biological processes. In order to describe the complex molecular motion of the environment in a simple manner, we introduce a system-bath model in which the intramolecular modes with anharmonic mode-mode couplings are described by a system Hamiltonian, while the other degrees of freedom, arising from the environmental molecules, are described by a heat bath. Employing a machine-learning-based approach, we determine not only the system parameters of the intramolecular modes but also the spectral distribution of the system-bath coupling to describe the intermolecular modes, using the atomic trajectories obtained from molecular dynamics (MD) simulations. The capabilities of the present approach are demonstrated for liquid water using MD trajectories calculated from the SPC/E model and the polarizable water model for intramolecular and intermolecular vibrational spectroscopies (POLI2VS) by determining the system parameters describing the symmetric-stretch, asymmetric-stretch, and bend modes with intramolecular interactions and the bath spectral distribution functions for each intramolecular mode representing the interaction with the intramolecular modes. From these results, we were able to elucidate the energy relaxation pathway between the intramolecular modes and the intermolecular modes in a nonintuitive manner.